Fetching contributors…
Cannot retrieve contributors at this time
2298 lines (1360 sloc) 68.3 KB
distcc-3.2 "Back in black" 2011-10-11
* IPv6 support (adshea, Benjamin R. Haskell, Bob Ham).
* Optional support for GSS-API authentication (Ian Baker at CERN)
* Added an environment variable DISTCC_SKIP_LOCAL_RETRY for skipping
the local retry in case of a remote compilation failure (Ryan Burns).
* Some constants that were previously hard-coded in the
sources are now configurable via environment variables:
* Fixed the following 20 issues <>:
30: 3.0 on OSX: ValueError: Expected absolute path, but got '(framework'.
33: distcc --scan-includes dumps core
34: Zeroconf fails if avahi supports IPv6
35: Include server internal error: '<class 'sre_constants.error'>: ('bogus escape (end of line)',)'
36: distcc-mon-gnome displays multiple rows for same host/slot
39: build with gcc 4.1 snprintf.h error "expected declaration specifiers"
42: distcc --show-hosts fails when using Avahi with IPv6 support and +zeroconf for distcc.
43: Remove duplicate hosts from zeroconf list
44: Failed tests on make check (gentoo gcc-4.3.3)
46: Add TAGS target to
49: make check fails on PreprocessAsm_Case
57: Fix warnings
58: PUMP mode can`t PARSE the HOSTS
60: CPlusPlus test case fails with GCC 4.4.1
61: Gdb tests fail in pumped mode on Ubuntu 9.10 amd64
66: pump mode fails to compile certain KDE-4 packages, finds/looks for incorrect headers
70: variable overlapping causing errors..
75: include_server documentation, distcc_fallback flag incorrect
81: distcc 3.1 will not compile correctly with current version of gcc
distcc-3.1 "Humming along" 2008-11-21
* New "--scan-includes" option to distcc. This shows which
headers distcc would send to the remote machine in pump mode,
without actually compiling anything. This provides a simple
interface to the include server.
* New "--disable-Werror" option to configure.
This disables the use of gcc's -Werror option when building
distcc, allowing compilation of distcc to succeed even if
gcc issues warnings.
* Document all of the exit codes in the man page
(Carsten Wolff <>, via Debian).
* Handle "-Wp,-MMD,...".
* Report a better error message if a host specification contains ",cpp"
but not ",lzo".
* Fix a bug where the "m4" directory wasn't included in the
source distribution tarball, which caused it to stop working
if you ran
* Fixed a bug where the pump script was hard-coding /etc/distcc/hosts
rather than using the prefix specified to configure.
* Fixed build errors on Gentoo Linux and some other systems
(Matsuu Takuto).
distcc-3.0 "The moment has arrived" 2008-08-06
* Minor improvements to the man pages.
distcc-3.0rc4 "Pump up the volume" 2008-08-04
* Fixed a bug in 3.0rc3 that broke pump mode for C++ compilations
that use -I options that refer to a subdirectory of one of the
default system include directories, e.g. -I/usr/include/foo.
distcc-3.0rc3 "Third time lucky?" 2008-07-31
* New "--without-avahi" option to configure (Lei Zhang).
* Fixed a problem where pump mode was falling back to local compilation
when compiling with -I options that refer to a subdirectory of one
of the default system include directories (Nils Klarlund, Fergus Henderson).
* Allow distcc to distribute commands that use "-MF<filename>"
rather than "-MF <filename>", and likewise for "-MT<target>".
* Fixed a build problem on ARM and Solaris systems.
* Fixed a build problem on some Linux systems reported by Daisuke Nishikawa.
* Fixed failure of "make install" on Cygwin.
distcc-3.0rc2 "Anti-aliasing." 2008-06-19
* Fallback to local compilation for "-march=native" or "mtune=native"
(Maks Verver).
* Fix for deadlock in pump mode.
distcc-3.0rc1 "My prices are insane." 2008-06-09
* New protocol, version 3, allowing for preprocessing on distcc
servers ("pump" mode).
* New component: an include server for incremental static analysis of include
dependencies. This software is written in Python 2.4.
* Support for getting hosts via avahi (Lennart Poettering)
* gcc debug option -d must be run locally because it produces
secondary files. Reported by Nadim Khemir.
* Improved checking for errors on socket connection. Patch
from Dan Kegel.
* "Pump" code tested under Linux only. Processing of object code
rewrites debug info (so server paths are replaced with client
paths), but this works for ELF code only.
* Fix failure of 'make maintainer-check' in non-English locales.
(Niklaus Giger)
* Locks taken through F_SETLK or lockf() must be explicitly released,
because closing the file does not do so.
(Sascha Demetrio)
* Use python2.4 if available.
* Execute compiler with correctly redirected output using win32 apis on
Cygwin/Mingw. Place temporary files using GetTempPath. (Alex Besogonov)
* Make distccmon-gnome link properly with gnu ld's new --as-needed
flag. (Ben Skeggs, Lisa Seelye)
* New web site: <>.
* New repository: <>.
* New maintainer: Fergus Henderson.
distcc-2.18.3 "Porsche safari" 2004-11-30
* Fix for incorrect handling of -x and -specs from Michal Welnicki.
* Fix false pass in tests for -x and -specs.
distcc-2.18.2 "Bees and lavender" 2004-11-12
* Fix problem in checking errno after connect(). Fix by Jean
Delvare and Dan Kegel.
* Fix prototype problem causing build failure on amd64. Patch
from Andreas Jochens. (Debian #280399, #280644).
distcc-2.18.1 "Reality-based computing" 2004-11-03
* Fix run-time warnings about gtk_tree_model_row_changed() when
built with recent versions of GTK+.
* gcc option -specs= must be run locally so that it can read the
local spec file. Reported by Chris Yeoh.
* Fix Solaris build problem reported by Dimitri Papadopoulos.
* Fix problem with reading hosts files in DOS CRLF format.
Reported by Sebastien Perochon.
distcc-2.18 "Billionaires on stilts" 2004-10-12
* Fix longjmp() bug that can cause distcc to crash when a timeout
* Fix compile-time problem with dcc_trace when configured with
--enable-rfc2553. Reported by Greg Earle.
* Fix warning about warn_unused_result for older versions of gcc.
* No timeout on local preprocessing or on compilation. No
timeouts on DNS lookups other than whatever is built in to the
DNS resolver. There are timeouts on network transmission and on
opening connections.
* Remove mmap() for receive of uncompressed data, to avoid various
* The --allow option is now mandatory for daemon mode. As a
security feature, distccd will refuse to run without an IP
access control list.
* conftest.* built by autoconf is always done locally, to make
configure tests faster and safer.
* Better description of use with ccache.
* Fix various warnings reported by Dimitri Papadopoulos.
* Fix Tru64 build problem reported by Gary V. Vaughan.
distcc-2.17 "Divers Alarums" 2004-07-31
* Performance improvements for distccmon-gnome.
* Added timeous for connection to servers, transmission of jobs
and remote compilation. If the timeout expires, the job is run
locally. This should make distcc cope better if some or all of
the servers are unreachable or failing.
* Also add timeouts in server to kick out stalled or disconnected
* If remote compilation fails because of a signal in a later phase
of the compilation causing an exit with code >=128, retry
compilation locally. Patch from Jakub Stachowski.
* Manpage fix from Jean Delvare.
* Cygwin portability fix for WCOREDUMP, reported by Eric Frias.
distcc-2.16 "Salt" 2004-07-08
* Fix bug that might cause IP-based access control rules not to be
interpreted correctly on 64-bit platforms. (CAN-2004-0601)
* Fix small memory leaks in distccmon-gnome and distccmon-text.
* Use a GNOME status bar with a grab handle for distccmon-gnome.
Patch from Nathan Fredrickson.
* Better detection of systems with a popt library installed but
no popt.h. Reported by Sean Kelly.
* Fix bug in reporting crashed daemon children that showed up on
systems without strsignal, such as Mac OS 10.2. Reported by
Benjamin Reed.
* Update to work on ia64 linux.
* Simplify dcc_log_child_exited.
* Include copyright and licence in --version output, as suggested
by GNU programs.
distcc-2.15 "feel good for ten seconds" 2004-07-06
* Fix crash bug in receiving LZO compressed data into a mmapped
file. Reported by Arkadiusz Miskiewicz.
* Remove redundant temporary file cleanup in non-forking mode.
* Client calculates and can display the overall throughput in
preprocessed bytes per second for remote jobs.
* The deprecated fork-after-exec mode controlled by the
--no-prefork option has been removed.
* mmap is no longer used for compressed data to simplify the code.
* Additional fixes to Red Hat / Fedora scripts, by Colins Walters.
* Daemon now chdirs to $TMPDIR at startup rather than /. This
allows dumping core, among other things.
distcc-2.14 "Lake Albina" 2004-05-02
* The host file is looked up in $DISTCC_DIR/hosts, not hardcoded
to ~/.distcc/hosts. The default for $DISTCC_DIR is still
~/.distcc. Suggested by Sebastien Perochon.
* Source file name is included in success/failure messages, e.g.
distcc[6655] ERROR: compile ./cases/bad.c on cardhu failed
* distcc can now be built with a build directory separate from the
source directory. Patch from Dennis Henriksen.
* Added --randomize option to host list to help spread the load
across large shared clusters. From patch by Google
(Josh Hyman <>).
* Add documentation of DISTCC_DIR, and other fixes. Patch from
Thomas Schwinge.
* Slightly improved distcc --help.
* Decompression buffer can dynamically resize to allow for very
gassy files. From a patch by Joe Buehler.
* Add sample scripts for Red Hat / Fedora, by Colin Walters.
* mmap is always disabled on HP-UX, because the inconsistent page
case is a bit dangerous for the way distcc uses mmap. Suggested
by Joe Buehler.
distcc-2.13 "Carnal Bunt" 2004-03-02
* Enforce IPv4 access control lists when the daemon is listening
on an IPv6 port. This only applies when the server was compiled
using --enable-rfc2553 and run with --allow.
* Hash comments are allowed in host list files.
* Use TCP_DEFER_ACCEPT on Linux to avoid some wasted context
switching when a new connection comes in to the server.
* Some fixes for connections over SSH when built using
--enable-rfc2553. Reported by Jeff Rizzo.
* Handle arguments like "-xcpp" and "-Wa,-xarch=v8". Suggested by
Ben Scarlet.
* Fix resource exhaustion when DISTCC_SAVE_TEMPS is set. Reported
by Mark DeGeorge.
* Fix problems in running IPv6-enabled builds on kernels without
IPv6 support. Reported by Lisa Seelye.
* More instructions for distributors/packagers.
* Many portability patches from Albert Chin. These patches were
tested on: AIX 4.3.2, 5.1; HP-UX 10.20, 11.00, 11i; IRIX 6.5;
Redhat Linux 7.1, 9; Solaris 2.5.1, 2.6, 7, 8, 9; Tru64 UNIX
4.0D, 5.1.
* Possibly fix mmap problem on HP-UX with compressed transfers.
Reported by Joe Buehler.
* Add test for compilation with compression.
* Testsuite portability patch from Albert Chin. The tests pass
on: HP-UX 10.20, 11.00, 11i; IRIX 6.5; Redhat Linux7.1, 9;
Solaris 7, 8, 9; Tru64 UNIX 4.0d, 5.1.
distcc-2.12.1 "Shatner's Bassoon" 2004-01-09
* When checking if a process exists, accept EPERM as indicating it
exists but isn't ours. This allows watching the directory of a
different user, subject only to filesystem permissions.
Reported by Ernst Bachmann.
* Correct a bug introduced in 2.12 that made distccd fail to start
when run by root because setgid failed. Fix from Wayne Davison.
* --without-gnome, --without-gtk now turn those options off rather
than on.
* New man page for distccmon-text, by Shri Shrikumar and Martin
distcc-2.12 "Figgy Pudding" 2003-12-19
* New --log-level feature allows logging of only errors rather
than messages for each connection.
* IPv6 support now works better when distcc is configured with
--enable-rfc2553. The server can normally accept either IPv4 or
IPv6 connections.
* New --with-docdir configure option to set install location for
documents. The default is the same as previously. Patch from
Harold L Hunt II.
* GNU-style ChangeLog is now included in the package. Please keep
it up to date if you make your own changes!
* setgroups() was broken by Mac OS X Panther so that it will not
accept setgroups(0, NULL) to clear the group list. Instead we
now set the group list to our single primary gid. Suggestion
from Wayne Davison.
* sendfile problems on FreeBSD 4.9-RC fixed. Patch from Allan
* MAXPATHLEN is provided if it's not defined, which is the case on
GNU Hurd. Reported by Santiago Vila as Debian #219600.
* Check pkg-config information GNOME/GTK+ is now checked at
configure time, not build time. This is a bit quicker than
running the script for every file we build, and means that
PKG_CONFIG_PATH only needs to be set for configure.
* Small documentation corrections.
distcc-2.11.2 "Muscovado" 2003-10-23
* Correct the handling of --without-gnome. Reported by Dimitri
* Fatal signal handlers for client and daemon should reset the
signal vector before doing anything else, to protect against an
(unlikely) spin if the signal is re-sent.
* Corrections to .desktop file from Dag Wieers.
* Remove dependency on libglade for GTK+ and GNOME monitor
* IRIX portability fixes from Dimitri Papadopoulos.
* Use AC_TYPE_SIGNAL to generate correct signal handler return
types. (It can be either int or void.)
* Various manual updates.
* Update to xinetd example from Dag Wieers.
distcc-2.11.1 "Burnt sugar" 2003-10-08
* The graphical monitor can now be configured with either
--with-gnome or --with-gtk. In the second case, the GNOME
libraries are not used. In both cases the resulting program is
called distccmon-gnome and the functional differences are fairly
* An error message is emitted and compilation fails if distcc
cannot use the specified TMPDIR. This can happen if distccd is
started by root with a TMPDIR only accessible to root.
Reported by Brett Dikeman.
* GNOME data files are now only installed when the application is
configured with --with-gnome or --with-gtk. Files are installed
into a distcc/ subdirectory of the configured data directory.
(By default this is /usr/local/share/distcc/). Reported by Erik
Jan Tromp and Noe`l Ko"the.
* Fix bug in sending compressed files where errors were ignored.
* C++ compilation using -frepo must be local. Reported by Andrew
* Don't use stdint.h, which is not available on FreeBSD 4-STABLE,
IRIX or Solaris. Reported by Frerich Raabe.
* Corrections to distcc manpage reported by Lionel Sausin.
* Documentation of how to write monitor programs, from Frerich
* Fix state.h for use with C++. Patch from Frerich Raabe.
distcc-2.11 "Sugar fix" 2003-09-23
* The GNOME monitor is now configured using --with-gnome (not
--enable-gnome) to be more consistent with GNU standards.
* The state file format has changed, so old distcc monitor
programs will not see new compiler processes and vice versa.
* Nice new GNOME monitor, showing strip-charts (or "scarves") of
the state of each processor slot over time.
* --allow is accepted as meaning "allow access from
anywhere". Debian #207831, suggested by Matthijs Kooijman.
* If we fail to determine the number of CPUs on the machine,
assume it is 1. This may fix problems on exotic machines where
the detection call is almost but not quite right.
* In non-forking mode, make sure to close the accepted fd after
processing a request.
* Use "safe" LZO1 decompression for protection against program or
network errors.
* Lock and state files are created with the weakest permissions
permitted by the umask. This is intended to reduce problems
seen when $DISTCC_DIR is shared between two users, such as when
compiling as root during program installation.
* Client deletes temporary files when interrupted.
* Add debugging option DISTCC_MMAP, which can be set to 0 to
disable use of mmap for receiving files.
* New --enable-profile configure option to turn on gprof.
* Many changes to the representation of the client state to make
it easier to draw a history. State is now stored as a natively
encoded struct in a disk file, rather than using our network
* Sleep only one second when blocked waiting for a CPU, to try to
reduce idle time when recovering from an overloaded period.
* Document problems with "no_subtree_check" on NFS.
* The GNU Free Documentation License is no longer distributed,
because the documentation is now included under the GNU GPL.
* Document bad Makefiles or mismatched compilers as common causes
of failure.
* Describe where to put "localhost" in the host list.
* Fixes for BSD/OS (BSDi) from Nick Amato.
* Fixes for portability issues in from Lisa Seelye and
Alexandre Oliva.
distcc-2.10.1 2003-08-12
* Fix problem in "make clean" reported by Lisa Seelye.
distcc-2.10 "Hunger is the best sauce" 2003-08-11
* Fix issues in temporary file handling that could allow a local
attacker to interfere with another user's distcc processes.
* Daemon niceness is incremented by +5 by default. (This can be
changed by --nice.)
* The per-user configuration directory can be set by $DISTCC_DIR.
It defaults to the old value of ~/.distcc/.
* For regular compile failure, don't say "with exit code 1".
* The server now emits warnings to the log and to the client if
its PATH is set such that it will recursively run a distcc
client. This check is done at run time for the requested
compiler name. The server's $PATH is not modified in other
ways. This may help reduce problems with finding the right
remote compiler.
* SCO/Caldera operating systems are no longer supported due to
their recent absurd attacks against Linux and IBM.
* TMPDIR is no longer required to start with '/', so that Cygwin
paths starting with a drive letter may be used. Reported by
Heiko Elger.
* Various man page updates.
* Daemon now sits in the root directory while running, not the
temporary directory.
* Temporary directory is gone. Working files are created directly
under $TMPDIR.
* Lock files and state files are stored under DISTCC_DIR, rather
than in /tmp.
* Test cases are more robust in choosing a daemon port number.
distcc-2.9 "Grace under pressure" 2003-07-21
* Optional LZO1X compression of network traffic, to make distcc
faster on slow or overloaded networks. This is turned on by
appending ",lzo" to each host specification.
Both client and server must be updated for this to work. Old
clients and servers are supported when compression is off, which
is the default.
Compression is probably only worthwhile on networks slower than
100Mbps, but it will depend on the usual factors of CPU, network
and code complexity. Thanks to Markus F.X.J. Oberhumer for the
LZO library.
* Handle invocations like "distcc foo.o -o foo", which can be
generated by Mozilla.
* Handle invocations like "distcc -Wa,-al=foo.lst -c foo.c", which
write an assembler listing to a file and cannot be run remotely.
* When trimming the path to avoid recursive invocations, stop as
soon as we find one non-symlink compiler. This should help
avoid the path getting trimmed down far too much on the server.
Patch from Wayne Davison.
* The daemon by default allows 2+NCPUS, rather than 3*NCPUS. The
two extras are to allow for some processes to be blocked on
network IO, but allowing more than two of them is probably not
useful on multiprocessor machines.
* Larger files are sent and received using memory mappings where
* Use of mmap and compression makes distcc substantially faster in
at least some cases.
* Convert the documentation to a series of regular Unix manpages
for easier reference and so they can be used more easily on
diverse platforms. The Linuxdoc-SGML manual has been removed
from the distribution. Thanks to Tim Potter for review
* Add tests for new -o and -Wa,-al handling.
distcc-2.8 "Play in the traffic" 2003-07-09
* Small enhancements to GNOME monitor appearance.
* Fix problem with nonblocking IO for SSH connections that causes
a busy spin. SSH connections are now far more efficient.
Reported by Marcelo Matus.
* distccmon-text flushes output after each line, even if stdout is
not a terminal. This allows it to more easily be run in polling
mode under another program. Suggestion from Hamish Rodda.
* distccd now works properly with SSH servers that use pipes
rather than socketpairs to run the daemon. Previously it would
fail with an error when trying to write to stdin. Suggestion
from Felix Lee.
* Correctly handle invocations like "gcc -ofoo.o -c foo.c", where
the output filename is run on to the -o option. Based on a
patch from Tsutomu Yasuda.
* mon.h is now safe for inclusion in C++ programs, and protected
against multiple inclusion. Suggestion from Frédérick Forjan.
* Increased timeout for nonblocking network IO apart from
connections to 5 minutes.
* Detect and avoid Apple gcc cpp-precomp brokenness. Patch from
Matt Watson.
* Better handling of builds on machines which do not have the
linuxdoc-sgml toolchain. Suggestion from Frerich Raabe.
* Include an example xinetd configuration file from Andrew Morton.
distcc-2.7.1 "Expedition" 2003-06-25
* distccmon-gnome now depends only on GTK+ 2.0, not on the GNOME
* Fix for compilation of the GNOME monitor on BSD.
* Fix for compilation of the monitor on Cygwin.
distcc-2.7 "The Hard Way" 2003-06-16
* Graphical compile monitor for GNOME 2.x. Configure with
--enable-gnome to build it.
* Possible fix for compilation on HP-UX.
* Sanity check GNOME libraries during configuration if
--enable-gnome was specified.
distcc-2.6 "The Watcher in the Water" 2003-06-12
* New program distccmon-text shows the state of any distcc
processes run by the current user. It can be run by hand or
read by a shell script. Should be popular with vmstat fans.
* New environment variable $UNCACHED_ERR_FD allows client errors
to be redirected from stderr to another file descriptor, so that
distcc errors are not cached by ccache. This is an improvment
because (almost?) all distcc messages are transient things such
as failure to connect to a remote host, and it is confusing to
have them cached.
(This needs a corresponding patch to ccache to set this, or for
it to be assigned to a duplicate of stderr before invoking
* In compilation log message, just show the basename of the file
to make it more compact.
* If an compiler cannot be executed then try searching the path
for the basename of the compiler. This can fix cases where the
compiler is invoked with an absolute name but it is in a
different directory on the server. Patch from Paul Russell.
* Fix build on platforms that don't have hstrerror(), such as
Solaris 2.6.
* Handle sysconf(_SC_NPROCESSORS_ONLN) returning 0 processors,
which has been observed on ARM Linux and previously threw
distccd into a spin. Patch from Sean MacLennan.
* New mon.c interface for writing compilation monitors.
* State directory name includes hostname for robustness when the
temporary directory is shared.
* Client writes out notes about its state to files that can be
watched by external monitor programs.
* Partial GNOME-based monitor interface, built when distcc is
configured with --enable-gnome.
* Fix silly typo that made maintainer-check give a false failure
in ParseHostSpec_Case.
distcc-2.5.1 "disable extra paranoia" 2003-06-05
* Fifteen second timeout on connection attempts. Hosts that are
unreachable in this time are blacklisted for another minute.
* Fix setpgid() failure when started directly from init(8).
Reported by George Garvey.
* Correct parsing of old-style TCP multiplier specification
(e.g. "phathost/10:3600"). Patch from Wayne Davison.
* Fix socklen_t compilation problem on Darwin.
* Fix compilation problems on Solaris 8. Report from Tomas Ögren.
* Fix compilation problems on IRIX. (IRIX cannot accept a
variable named "sa_len".) Report from Peter Naulls.
* Fix compilation on SuSE Linux 7.1 for Alpha.
* Add more details about starting distccd to the user manual.
* Add test for correct parsing of old-style TCP multiplier
specification (e.g. "phathost/10:3600").
distcc-2.5 "give them heaps" 2003-05-27
* The standalone daemon now "preforks" before accepting
connections to reduce per-request overhead. Can be disabled by
--no-prefork, which restores the standard Unix behaviour.
* Restrict the number of jobs accepted onto the server at any
time. By default this is determined automatically depending on
the number of CPUs on the server, but it can be set using the
--jobs, -j option. Replaces never-used --concurrent, -n
* When the daemon is terminated, the entire process group is shut
down, including any compilations in progress.
* Load limits on TCP hosts can now be specified in the more
consistent form of "HOST:PORT/MUL" rather than "HOST/MUL:PORT".
The old form is still supported.
* Better error message for "bind failed".
* Better "listening on" message at startup.
* More meaningful reporting of transmission and compilation times
in logs.
* Fix bug that caused some messages to go to the 'user' syslog
facility rather than 'daemon' as presently.
* Fix problem with terminating the daemon when running under
* Refactor daemon loop.
* Server-side temporary object files are called .o, not .out.
Suggestion from rishikesh shetty.
* Temporary directory name changed to decimal userid, rather than
hexadecimal userid. Easier to recognize.
* Daemon changes into state directory at startup.
* SSH child is collected and checked when the connection is
distcc-2.4.2 "hard and fast" 2003-05-23
* Fix compilation directed to /dev/null by handling output files
the same way as the GNU BFD library. Reported by Hal Duston.
* Better error messages if DISTCC_HOSTS or host files are empty.
Patch from Wayne Davison.
* Fix string allocation bug in dcc_sockaddr_to_ip that could cause
a server crash.
distcc-2.4.1 2003-05-22
* Add strndup() for non-GNU platforms. Patch from Frerich Raabe.
distcc-2.4 "sparky will reign" 2003-05-21
* Preliminary support for distribution of Objective C and
Objective C++ compilation. Based on a patch from Dara Hazeghi.
* If remote compilation fails, give a visible error. People might
want to know, particularly if something is failing on one
particular machine.
* Error messages report host definitions more consistently, rather
than just hostnames.
* Better handling of attempts to run more concurrent compilations
than there are available slots. distcc now sleeps for a few
seconds to avoid thrashing the client machine.
* Unavailable servers are now shunned for 60 seconds so as to
avoid wasting time connecting to machines that are down, not
running distccd, etc. (However hosts that are simply not
responding (e.g. firewalled) are not handled perfectly yet.)
* If an IO error occurs while trying to get a lock, bail out rather
than getting stuck.
* Don't use the RFC2553 getaddrinfo/getnameinfo API by default,
because it seems to be slower on OS X and possibly problematic
on other systems. It can be enabled with --enable-rfc2553 if
you want to use IPv6, multi-homed servers, or want to test the
new API.
* Preliminary STRATUS VOS portability work from Paul Green.
* More description of execution over SSH.
distcc-2.3 "come down like a ton of bricks" 2003-05-16
* Respond more gracefully to protocol derailments. In particular,
banner messages sometimes seem to get into the stream when
distcc is run from inetd, and they were previously a bit hard to
diagnose. This won't fix the configuration problem that causes
them, but it should make the problem obvious.
* The client now tries all addresses for each server, in the order
returned by the resolver.
* Client and server now use IPv6 if requested, where this is
supported by the system. The server still binds to an IPv4
address by default, but this can be set by the --listen option.
Client access control is not supported yet for IPv6.
* Fix possible bug in loading host list files.
* Startup errors are now reported to stderr, unless that is a
socket in which case they are sent to syslog. This should be a
better tradeoff between making errors obvious and the
requirements of inetd mode.
* Add description of --listen to manual, plus other improvements.
Patch from Frerich Raabe.
* Various clarifications to the manual.
* Add documentation of DISTCC_FALLBACK, DISTCC_SSH, execution over
SSH, and host list files.
* Two Mac OS X fixes from Benjamin Reed: Use the BSD method to
discover the number of CPUs, and include types.h where it is
* Improved portability of snprintf fallback implementation. Now
works again on dietlibc/Linux. Some fixes merged from Samba
distcc-2.2 "nosurname" 2003-05-05
* If $DISTCC_HOSTS is not set, the host list is taken from from
~/.distcc/hosts, if that exists, or otherwise
${sysconfdir}/distcc/hosts. (The location of this last one is
set by the configure option --sysconfdir. It defaults to
* Add --listen option to distccd, to control which IP address is
used to listen for connections. May be useful for access
control on dual-homed machines. Based on a patch by Neil
* Add documentation for --allow and --user, by Frerich Raabe.
* Add test case for host list file.
* Try to prevent server messages getting into syslog during
distcc-2.1 "Don't mind if I do!" 2003-05-03
* Support for execution over ssh!
Hosts to be contacted over SSH should be specified as '@HOST' or
'USER@HOST'. distccd must be installed on the volunteer host, but it
need not be running and there is no need to set up ssh tunnels.
Compilations run under the user's account, and connections are securely
authenticated and encrypted.
SSH is substantially slower than plain TCP connections, but they can
still be faster than local compilation.
* The search path is examined in both the client and server, and
directories containing distcc (and ccache, where appropriate) masquerade
hooks are removed. This should help prevent distcc accidentally
invoking itself recursively on either client or server. Patch from
Wayne Davison.
* Fix up warning about pid file being removed twice.
* Better error handling when corks are not supported on a particular
system or socket.
* Fix incorrect display of CPU usage statistics in server log and client.
* Change behaviour of distccd for log messages encountered during startup,
before options are parsed. These are now written to stderr only if that
is a tty, and otherwise to syslog. This is intended to prevent log
messages from distccd getting mixed into the network protocol when it is
run from inetd.
* Correctly trap unexpected end of input when copying a file using
* Test harnesses may need snprintf.o linked in as well. Reported by Joe
Meslovich as broken on Solaris.
* Add test that --allow does in fact deny connections as it should.
* Add tests for parsing of --allow options.
* Add DISTCC_FALLBACK environment variable. If set to 0, means that
failure to distribute will be considered an overall failure. By
default, a networking error means that compilation will be retried
* Many cleanups of error handling code.
* Trace distcc version at startup to help with bug reporting.
* Update IO code to handle nonblocking sockets, because we (may?) need to
use nonblocking socketpairs to talk to SSH.
distcc-2.0.1 "colorful savings on all your favorite medications" 2003-03-28
* Fix library missing from distribution.
distcc-2.0 "shoot through like a Bondi tram" 2003-03-28
* Fix small signal handling race that would occasionally cause distccd not
to exit when signalled. Pointed out by Vance Lankhaar.
* If DISTCC_LOG is an empty string, treat as undefined.
* If the file specified by DISTCC_LOG can't be opened, then show an
error and use stderr. Previously, error messages were just lost.
* New "masquerade" mode: distcc can be linked into a directory on the
search path as 'cc'. By Wayne Davison and Martin Pool.
* Detect number of online processors on the server. At the moment it is
just logged but in the future it might be used to guide scheduling.
Thanks to Dimitri Papadopoulos-Orfanos for describing how to do this on
many operating systems.
* Log signal name when terminating, on systems with strsignal().
* Internal CPU efficiency fixes in logging and RPC code.
* Apparently now builds on IRIX. Reports about whether it works would be
* Fix cast in call to sprintf for platforms where ssize_t != int. Found
on Alpha Linux in the hp Compaq testdrive.
* Clean up some warnings on IRIX.
* Fix problem with vsnprintf() declarations on Solaris 8.
* Cope when in_addr_t and socklen_t are not defined, as is the case on
FreeBSD 3.3.
* Fix test suite to allow for 'true' and 'false' being in /usr/bin on
* Many internal updates to ComfyChair framework.
* Test suite now works with Python 1.5 or later.
* Add new undocumented (heh) "--lifetime" option that causes the daemon to
expire after a certain period.
* Specify limited lifetime when running daemon, so that if it is not
cleaned up by the tests it will go away after a little while anyhow.
* Use "--allow" to limit connections to localhost when running
daemon for testing.
* Use randomly chosen TCP port for the server when running tests. This
makes repeated or concurrent invocations of the test less likely to
* On machines without Python, just say "test skipped" rather than failing.
* Test DISTCC_LOG being set to an unwriteable file.
* The GNU Info form of the manual is no longer included in the source
distribution or installed by default. It can be built from SGML if
desired, with "make install-linuxdoc-info".
distcc-1.2.3 "stubborn tiny lights" 2003-03-02
* Don't use ':' in lock filenames, because that's a bit too tough for
Windows. This only affects Cygwin clients, not servers. Pointed out by
Heiko Elger.
* Check existence of ctype.h. Fixes warnings when building popt. Pointed
out by Tom Matelich.
* Fix "access denied" errors on lock files on HP-UX and Cygwin, so clients
on those platforms should work. Reported by Tom Matelich, Heiko Elger
and others.
distcc-1.2.2 "clustering darkness forever" 2003-02-27
* "make distclean" should not remove ./configure.
* Don't assume POSIX-style reentrant file locks, which can't be supported on
* Check for inet_aton in libresolv, which might fix Solaris 2.6.
* Remove some gnudity from the Makefile in an attempt to build with
non-GNU Make.
* Attempted fix for header files on BSD.
* Attempted sendfile implementation for HP-UX.
* Add a simple "make distcheck" target to test that the tarball is built
distcc-1.2.1 "you know you're right" 2003-02-25
* Better detection of python2.2 during configuration.
* Allow installation from CVS on machines that don't have the Linuxdoc tools.
* Check for autoconf >= 2.53 when configuring. Remove dependencies on
features from later versions of autoconf to allow autogen to run on Red
Hat Linux 7.2 systems.
* Add "make installcheck" target.
* "make check" now fails if any of the tests fail to run.
* distcc is now built on the build farm. Please contribute
machines if you want your platform to be properly supported.
* Remove libiberty, and use Samba's asprintf() implementation instead. Yay
copy and paste. This should fix problems with asprintf and ansidecls.h on
Solaris, BSD, and Cygwin.
distcc-1.2 "lizard drinking" 2003-02-22
* New scheduler makes a better attempt at keeping remote machines solidly
loaded by allowing only one request at a time to be in transit to remote
machines, therefore letting the actual compiler start sooner. Gives
about a 10% improvement in some circumstances. The new algorithm should
also be a better foundation for adding compression or server-side load
* New --allow option for distccd allows IP-based access control in either
inetd or standalone daemon modes.
* Locking mechanism now distinguishes between daemons on different ports
of the same machine for the purposes of load balancing.
* Print timing and estimated rate on bulk data transfers.
* Open binary files in O_BINARY mode on Cygwin. May fix some corruption
problems on this platform. More investigation would be appreciated.
Based on a patch by Helind.
* Use putenv() rather than setenv() to fit into SUSv2 and work on Solaris
8. Pointed out by Dimitri Papadopoulos.
* Add some functions from libiberty to aid portability.
distcc-1.1 "balls to the wall" 2003-01-28
* Flag platforms that have a sendfile() function, but that aren't
explicitly supported, because it seems to vary substantially. At the
moment only FreeBSD and Linux are known.
* Fix configure test for hstrerror(). Patch from Frerich Raabe.
* Check for inadvertent recursion, where distcc calls itself as the
compiler. Causes an exit with value 111.
* Don't log client hostname, only the IP address. It's not really
necessary and can slow us down.
* If run by root, discard privileges rather than flagging an error. The
persona to assume is set by the --user option and defaults to "distcc".
The user "nobody" or uid 65534 is used as a fallback if the named user
is not found.
* Allow for running benchmarks repeatedly, printing the mean and standard
deviation of run time.
* Projects may now be selected by an unambiguous prefix of their name, so
you can just say "benchmark hello" rather than "benchmark hello-2.1.1".
* Print better explanatory messages while the benchmark is running.
* Update some tests to work with changed behaviour of gcc 3.2 as well as
* Many updates to distcc User Manual.
* Comply with GNU Standards by keeping all messages in the NEWS file, not
just those for the current release.
distcc-1.0 "consider the oyster" 2003-01-16
* Fix sendfile() bug on FreeBSD that could cause files to be
mistransmitted. Perhaps not manifest on x86? Reported by Kris
* Distribute as .tar.bz2 rather than .tar.gz. (We pay by the byte.)
* The recommended convention for gcc names is <target>-gcc-<version>, and
GCC 3.3 will install itself under this name. (Alexandre Oliva, Claes
* New macro-benchmark and test suite for distcc. Invoke it with "make
benchmark." It takes a long time to run and causes a lot of source code
to be downloaded, but should be self-contained.
This is the start of a script that runs distcc to build various open
source programs, as an overall test of correctness, and to measure
performance of a distcc installation. It relies on the user to install
and configure distcc servers and clients, but handles the work of
downloading, preparing, and building packages by itself.
There are still some rough edges on the script, but it may be of
* Add control over which phases of the benchmark are run.
* Add ability to run tests repeatedly and report mean and standard
* Remove Plucker file: nobody seems to be downloading it, and it's
easy enough for people to build if they're interested.
* Remove PDF and Postscript from documentation in tarball. They are still
available from the web site or may be built from source.
distcc-0.15 "petrified forest" 2002-12-12
* Change default port from 4200 to IANA-assigned standard for distcc
service "3632". Compatibility with old installations can be retained by
explicitly setting --port=4200 on the server, or using "HOST:PORT" on
the client.
* Fake-root directory for "make install" is now set by DESTDIR, not
TARGET_ROOT. (Alexandre Oliva.)
* Run the preprocessor with output sent to stdout and redirected to a
file, rather than by using the -o option. This should fix some
compilers that can't handle "-E -o". It may improve gcc behaviour with
-MD, but because gcc behaviour has changed over time it may not be a
complete fix.
For gcc 3.2 it is recommended that -MF be used to explicitly set the
dependency output filename.
* Add test case for transmission of a zero-byte file.
* Add badly-behaved compiler drop-in for use in testing.
* Use the correct autoconf macros to detect whether install(1) is
available and working, and if not, fall back to using the included
install-sh script. This should fix "make install" on Solaris. (Mike
* Fix waitpid() takes three options. Required for Irix. (Michael Santy)
* Fix autoconf detection of resolv.h on Solaris. (Dimitri Papadopoulos)
* Describe information to include in a bug report.
* Remove "Results" chapter from manual: there are now many more detailed
results on the web site, and this information is better suited to
storage there, because it will be updated on a different cycle.
End-users don't really need to have it in their reference.
* Remove "MOC Compatibility" section from manual because at the moment
there is nothing to say.
* Add a new section about the complexities of gcc -MD.
* Remove "Bugs" section from the manual, and put it into a new TOOD file
instead. Keep information about reporting bugs and using the test
suite, but put it in a different section.
distcc-0.14 "no agonizing hanging weights" 2002-11-13
* Take lock on localhost only when compiling locally, not for all
compilations. This removes the accidental limit of only 4 overall jobs.
(Martin Pool)
* Fix Makefile bug that caused HTML documentation to be missing from the
distribution tarball. (Frerich Raabe, Martin Pool)
* Make PreprocessPlainText_Case run in appropriate subdirectory. (Martin
* Test for wait4 and wait3, which are missing on IRIX. (Mike Santy,
Martin Pool)
* Include example init.d script. (Jason Thomas)
distcc-0.13 "Armistice" 2002-11-11
* New --no-detach option to cause the server not to detach from its
parent. It still forks normally when a connection is accepted. This
may be useful with daemontools or other frameworks that prefer the
daemon not to detach. (Martin Pool)
* Log messages on the server of severity "warning" or higher are captured
and sent back to the client. (Martin Pool)
* If the local preprocessor fails, do not re-run the job locally because
it would presumably fail in the same way. (Martin Pool)
* By default, the client only sends "warning" or higher messages to
stderr, so that nothing extra is logged unless there is possibly a
failure in distcc itself. With a DISTCC_LOG file, "notices" and higher
are logged. (Martin Pool)
* Better messages when a compiler exits or crashes. (Martin Pool)
* If the compiler because of a signal, return 128+SIGNAL, per Unix
convention. (Martin Pool)
* Compilers and preprocessors are now run in their own process groups, and
terminated if the daemon is killed. (Martin Pool)
* When the daemon parent is terminated, it does not kill all its children.
Instead, they're allowed to finish whatever they're working on, and then
exit. This allows for more graceful shutdown. (Martin Pool)
* More specific exit codes for some error cases. Don't panic. (Martin
* Include popt library and --with-included-popt configure option to aid
installation on systems that don't have it. By default it is statically
linked in only if the build machine does not have a suitable library and
header. (Martin Pool)
* Remove support for feeding the compiler from a fifo. The gain from
using fifos does not seem to justify the maintenance burden. (Martin Pool)
- This makes the behaviour more consistent, because some platforms
(Cygwin) or filesystems (NFS, strangely) can't use fifos. It removes
a code path and a slightly complex autodetection.
- Using fifos makes some potential scheduling improvements hard: the
server can't make good decisions about how many tasks to run, because
each one will only use a fraction of the CPU. The client cannot
serialize file transmission, which would probably be desirable.
- Using fifos makes reliably handling compiler failures slightly
harder: the compiler may crash or exit at any time, which will in
turn cause the server to hang if it later tries to open or write to
the fifo. In previous versions, distcc tried to handle this by
catching SIGCHLD when the child terminated, and aborting the
operation, but I am not sure that the method is completely reliable.
* Detached daemon now ignores hangup signal. (Martin Pool)
* When the parent is terminated, don't kill its process group. Allow
children to finish in their own time. (Martin Pool)
* Detect the case where cpp claims to have succeeded, but did not actually
produce any output. (Martin Pool)
* Do not use atexit() to clean up temporary files, because this can cause
havoc if it's ever called from a child process that forked but failed to
exec. (Martin Pool)
* Handle "gcc -S foo.c -o -", which ought to write assembly to stdout.
(Alexandre Oliva)
* Fix bug in handling IO errors (e.g. dropped connection) in the server.
(Martin Pool)
* If we fail to cork a socket, continue anyhow. (Martin Pool)
* Make error handling for IO more consistent. (Martin Pool)
* Follow Unix convention of returning 128+SIGNAL if the compiler exits
with a signal. (Martin Pool)
* Add tests for parsing slot limit from host specification. (Martin Pool)
* Daemon tests collect dead servers faster, so the tests run a bit
faster. (Martin Pool)
* Nicer handling of ^C while running tests: print "INTERRUPT" and
terminate the whole test. (Martin Pool)
* Add new tests for scanning command lines with an implied compiler name.
(Martin Pool)
* When the parent is terminated, don't kill its process group. Allow
children to finish in their own time. (Martin Pool)
* Detect the case where cpp claims to have succeeded, but did not actually
produce any output. (Martin Pool)
* Add test case for #error directive. (Martin Pool)
* Add test case for preprocessing non-C text, as is done by (for example)
xrdb. (Martin Pool)
* Add test case for invocations like "distcc -c test1.c test2.c". (Martin Pool)
* Add test for handling of a compiler missing from the server. (Martin Pool)
* Allow distcc's version to be set to something like "0.13cvs-nofifo".
(Martin Pool)
* Add test for handling of a compiler missing from the server. (Martin
* Improved ComfyChair testing framework. (Tim Potter, Martin Pool)
* Includes example init.d script for SuSE 7.x and 8.0. Not installed by
default. (Brandon Forehand)
* Document use with Gentoo Linux. (Dean Bailey, Ernesto, Martin Pool)
* Get rid of recursive Make and fix various small Makefile bugs. (Martin
* Add $(TARGET_ROOT) variable for "make install", for use in building
binary packages. (Martin Pool, Nick Moffitt)
* wait4() on Solaris can't handle a pid of -1, which means "any child" on
Linux and BSD. Use pid of 0 instead to collect any children from the
same process group. (Kevin Bailey)
* Client does not abruptly drop network connection on compiler failure but
rather reads the 0-byte object file token. (Martin Pool)
* Rework logger library to allow multiple logger callbacks to be active at
once. This is used for the server, which can write to its own log files
and also send a copy of messages to a buffer to be sent to the client.
(Martin Pool)
distcc-0.12 "Klein, kosten- und schnörkellos" 2002-10-07
* Handle "gcc -S foo.c -o -", which ought to write assembly to stdout.
(Alexandre Oliva)
* wait4() on Solaris can't handle a pid of -1, which means "any child" on
Linux and BSD. Use pid of 0 instead to collect any children from the
same process group. (Kevin Bailey)
* Add tests for "-o -". (Martin Pool)
distcc-0.11 "Nuclear Bedtime Story" 2002-09-25
* More helpful message explaining that popt is needed if it is missing at
build time. (Martin Pool)
* Better stripping of options from remote command lines. (Martin Pool)
* Handle the new -MM, -MG, -MP, -MF, -MT, -MQ options for gcc 3.x.
(Stephan Kulow, Martin Pool)
* Whenever a command by either the client or the server, it counts towards
the load on that machine. This covers undistributed commands, cpp, and
compilation. This should avoid the bias in earlier versions towards
overloading the client. (Martin Pool)
* IO errors should not cause the process to terminate straight away,
because they are properly handled and we may have important cleanup to
do. (Martin Pool)
* Fix error handling in code for executing compilers. (Martin Pool)
* Fix leakage of file descriptors when trying to acquire a lock. (Martin
* Do not object to running as group 0, because on BSD this is "wheel" and
many non-privileged users are in it. (Frerich Raabe, Martin Pool)
* Fix (non-exploitable) buffer overrun bug. (Dave Zarzycki)
* Fixes to build on Linux libc5. (Richard Zidlicky, Martin Pool)
* Keep $TMPDIR separate for each test case. (Martin Pool)
* now optionally takes arguments saying which tests to run.
(Martin Pool)
* Test permissions on output files are correct. (Martin Pool)
* Add more test cases for implicit compiler name handling. Extend
h_scanargs to exercise this. (Martin Pool)
* More tests for argument stripping. (Martin Pool)
distcc-0.10 "shiny fast red thing" 2002-09-12
* Allow distcc to be directly used with the same syntax as a compiler:
"distcc -c hello.c". At the moment, always calls "cc". (Martin Pool)
* If distribution fails, show the name of the server that we were trying
to use. (Martin Pool)
* Strip -D, -I, -L, -M* options before passing the command line across the
network, for clarity (and perhaps a tiny performance improvement.)
Also, this allows new clients which support -MD to be used with old
servers that can't handle it. (Martin Pool)
* Correctly handle "gcc -c -S hello.c", which should only assemble, not
compile. (Alexandre Oliva)
* If the volunteer compiler exits without reading all of its input, return
the compiler's own exit code across the network, rather than
* Attempt to fix a race where the compiler process could exit without
opening its input before we started writing to the FIFO. In that case,
the daemon would hang forever waiting for the compiler to arrive. In
the new code, SIGCHLD causes a longjmp to break out of trying to feed
the compiler. (Martin Pool)
* If .o file is 0 bytes or not created on the server, do not create it on
the client either. (Martin Pool)
* Fix bug that caused compilation to fail if an old compiler output file
existed. (Martin Pool)
* Try to make sure to uncork client socket before closing if cpp fails, to
work around Linux 2.2 bug that causes the socket to jam in FIN_WAIT1.
See <>.
(Martin Pool)
* Invocations of gcc that use -MD or -MMD can be distributed, although -M
cannot be. Both client and server must be updated to make this work.
(Martin Pool)
* It turns out that .s and .S files in fact cannot be assembled remotely,
because they might use the gas '.include' pseudo op, which is resolved
by the assembler and would end up reading something from the volunteer's
disk. (Richard Zidlicky, Martin Pool)
* Solaris portability fixes. (Dimitri PAPADOPOULOS-ORFANOS)
* Test handling of -c and -S. (Martin Pool)
* Test handling of a file that causes a preprocessor error. (Martin Pool)
* Test compiler that succeeds without reading input. (Martin Pool)
* Test many simple compilations, in case there is an infrequent error.
(Martin Pool)
* Test many concurrent compilations. (Martin Pool)
* Test using -D on the compiler command line. (Martin Pool)
* Tests for .s and .S files. (Martin Pool)
distcc-0.9 "samui desu ne?" 2002-09-03
* Add new environment variable for client and server, DISTCC_TCP_CORK.
Defaults to on, intended mostly for performance comparison or
debugging. (Martin Pool)
* Properly support platforms with 32-bit pids, such as Cygwin. (Aaron
Lehmann, Martin Pool)
* On FreeBSD, the sa_family_t type is defined in <sys/socket.h> (Dennis
* Fix bugs related to leftover fifos in the temporary directory causing
occasional server hangs. (Martin Pool)
* Stop the compiler from inheriting file descriptors connected to the
network. This was causing the client to hang if the server abnormally
terminated while the compiler was reading from a fifo. (Martin Pool)
* Handle the case of sendfile() transmitting only part of the supplied
data, similarly to a short write on a socket. This doesn't often
happen, but can occur if a ptrace debugger attaches while sendfile is
running. (Martin Pool)
* When run with --daemon and not --no-fork, the parent does not detach
until the pid file has been created. This removes a slight race window
that sometimes confused the test scripts. (Martin Pool)
* Message for completion of local cpp now gives the correct hostname.
(Martin Pool)
* Do not try to collect daemon children if --no-fork was specified.
(Martin Pool)
* Set SIGPIPE handler back to default before starting compiler tasks.
(Martin Pool)
* If a network error is detected while feeding input to the compiler, then
it is terminated with a SIGHUP. (Martin Pool)
* If transmission to the server fails, close the socket sooner rather than
later so as not to make the server process wait around unnecessarily.
(Martin Pool)
* Better handling of the case where the compiler command exits without
reading its input. (Martin Pool)
* Make sure compilation children are always properly collected, even if
transmission failed. (Martin Pool)
* Write token/parameters using a single write() call. May be very
slightly more efficient in CPU and perhaps packets. (Martin Pool)
* Display test names while they're running so that long tests are easier
to follow. (Martin Pool)
* When cleaning up after running the dameon, make sure it's really dead
before proceeding, so that tests don't trip over each other. (Martin
* New test of compilation of a largish (~4MB) C file to a ~5MB object
file. (Martin Pool)
distcc-0.8 "cat and girl" 2002-08-15
* If for any reason a job can't be distributed, it is run locally rather
than failing. This includes servers which are not running or
unreachable, servers which crash, bad host specifications, etc. (Martin
* --help for distcc explains more about host specifications. (Martin
* Correctly handle compilation of C++ code under ccache, by properly
recognizing the .ii extension. (Stephen White, Chris Halls)
* Boolean environment variables (e.g. DISTCC_SAVE_TEMPS, DISTCC_VERBOSE)
are now consistently required to be literally "1" to count as true.
(Martin Pool)
* Document that $DISTCC_HOSTS can now specify ports. (Martin Pool)
* Document --no-fifo and --log-stderr. (Martin Pool)
distcc-0.7 "Pipeline Authority" 2002-08-05
* Can now either use fifos for feeding the compiler or not. This can be
controlled by --no-fifo to the server. By default, it tries a fifo and
falls back to using regular temporarily files if that fails, as it will
on Cygwin or an NFS-mounted /tmp. (Martin Pool)
* --log-stderr option, intended mainly for debugging or testing. (Martin
* DISTCC_HOSTS can now specify non-default port numbers using "host:port"
syntax. (Martin Pool)
* Built documentation (HTML, PS, PDF, Info) is now shipped in the
distributed tarball, so that you can more easily install it on a machine
lacking the Linuxdoc SGML tools. You still need them to modify and
rebuild the documentation, of course. (Martin Pool)
* Log pid even if running with --no-fork. Remove pid file on exit.
(Martin Pool)
* Try fcntl locks in addition to lockf and flock; apparently this helps
portability to Cygwin. (Martin Pool, Marco Alanen)
* Port to Cygwin. (Martin Pool, Marco Alanen)
* Partial port to Solaris. (Dimitri PAPADOPOULOS-ORFANOS, Martin Pool)
* Further internal improvements to test suite. (Martin Pool)
* Test suite should now run whether or not there is a daemon listening on
localhost, because it uses a non-standard port for testing. (Martin
distcc-0.6 "reptilian agenda" (12 July 2002)
* Daemon now refuses to run as root, because this is just too ridiculously
insecure. (Martin Pool)
* Output shows CPU time taken by distcc and programs that it runs. (Martin
* distcc will now distribute jobs which use -s to compile but not assemble.
Previously they were always run locally, but there's no strong reason why
they must be. Unfortunately you need to upgrade both the client and
server for this to work, because old servers will refuse to run gcc with
-s. (Martin Pool)
* -M<anything> option implies that cpp must be run locally, because it will
want to write dependency information to a file. (Martin Pool)
* Add --daemon and --inetd options, so that you can properly start a remote
daemon with a single-line ssh command. (Martin Pool)
* Fix bugs related to collecting child processes. (Probably not very
user-visible.) (Martin Pool)
* Fixed some Makefile-related bugs causing "make install" breakage and
similar things. (Martin Pool)
* If the system supports sendfile, but the particular filesystem we're on
(e.g. tmpfs) doesn't, then fall back to using read/write. (Eivind Magnus
Hvidevold, Martin Pool)
* Change to argument analyzer to placate Valgrind. Possibly this was a
memory corruption bug, but I can't see it. I think Valgrind was just too
picky. (Martin Pool)
* Build on FreeBSD, though you do need GNU make. (Lauri Watts, Martin
* Better description of how to do cross-compilation. (Martin Pool)
* Improved Python test suite. (Martin Pool)
* Get rid of automake; just use plain autoconf 2.53. (Martin Pool)
* If $DISTCC_SAVE_TEMPS is set to "1", then files are not actually deleted
-- good for debugging. (Martin Pool)
distcc-0.5 "Unacceptable Behaviour" (27 June 2002)
* Use more specific return codes when the remote compiler fails:
EXIT_COMPILER_CRASHED for a signal, or the return code from the
compiler if it exited non-0.
* Fix bug in h_hosts that caused us to write off the end of the
real argv[], rather than a copy. Thanks to Julian Seward.
(Martin Pool)
* Cope on compilers without varargs macros. (Petter Reinholdtsen, Martin
* Build on Solaris. (Petter Reinholdtsen, Martin Pool)
* FreeBSD fixes. (Claes Wallin, Frerich Raabe, Michael
Nottebrock, Martin Pool)
* Manpages are correctly distributed and installed. (Petter Reinholdtsen)
* Test suite now runs a server on localhost and tests against it,
and shuts it down on completion. (Martin Pool)
* More test cases. (Martin Pool)
distcc-0.4 "Wisdom Like Silence" (9 June 2002)
* Assembly and assembly-with-preprocessor files can now be
assembled remotely. (Martin Pool)
* Client now emits no messages unless there is a warning or error,
or $DISTCC_VERBOSE is specified. This improves operation with
ccache. (Martin Pool)
* Client and server both respect $TMPDIR if set. (Martin Pool)
* Standalone server puts itself in the background and detaches from
controlling tty. (Martin Pool)
* Server has --port option. Currently useless, since there's no
corresponding option on the client. (Martin Pool)
* Server has --pid-file option. (Martin Pool)
* Server now allows multiple concurrent compile tasks, forking per
connection. (Martin Pool)
* gcc invocations with -fprofile-arcs or -ftest-coverage that
would emit coverage information to a local file are always run
locally. (Martin Pool)
* Handle lines like "gcc -c -c hello.c". (Martin Pool)
* Various enhancements to the user manual. (Martin Pool)
* Add a partial test suite based on PyUnit. Python 2 is now
required to run the test suite, though not to just build or run
distcc. (Martin Pool)
distcc-0.3 "The Anticipation of a New Lover's Arrival" (28 May 2002)
* Support C++ compilation (including with ccache) by correctly
detecting prefixes. (Ian Reinhart Geiser, Martin Pool)
* Use a FIFO to feed preprocessed source from the daemon into the
compiler, so that compilation can be overlapped with network
transit. (Martin Pool)
* Only fiddle with the compiler's stdin/out/err when running on
the server, not on the client. This should make cpp from stdin
work. (Reported by Ian Reinhart Geiser from KDE) (Martin Pool)
* Prevent server crashing on aborted connectiong by using EPIPE
rather than SIGPIPE in daemon as well as client. (Martin Pool,
Ben Elliston.)
* Builds on GNU/Linux, FreeBSD, and others if you're lucky.
(Martin Pool, Frerich Raabe, Luke Gorrie)
* Convert to using GNU automake and autoconf, so that distcc can
better handle portability, distribution and testing. (Martin
Pool, Frerich Raabe, Ian Reinhart Geiser)
* Start adding some "make check" tests. (Martin Pool)
distcc-0.2 (17 May 2002)
* First public release
* Correctly and usefully builds various large programs
Local variables:
mode: indented-text
indented-tabs-mode: nil
coding: utf-8